Landslidev : a visual analitics system for model comparison in landslide prediction

Abstract

Landslide and debris flows are hazards that can result in huge victims and economic losses. Landslide prediction is one of the most important approaches to mitigate these hazards’ effects, in which selecting suitable models for prediction is crucial. However, current model selection heavily depends on traditional evaluation metrics of the entire dataset, which are ineffective and inefficient due to the data complexity in the temporal and spatial analysis, and prediction priority among periods and locations. This thesis proposes LandslideV, a visual analytics system to interactively facilitate the model comparison for landslide prediction. It provides a comprehensive comparison among models using different features in various periods and regions. Besides, it supports further model optimization based on oversampling subsets of data of poor performance or high importance. Case studies with the real-world dataset are conducted to evaluate the effectiveness and applicability of LandslideV.

Date of Award	2022
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Huamin QU (Supervisor) & Charles Wang Wai NG (Supervisor)